An iterative dealiasing method based on mixed norm self-supervised learning

By adopting an iterative inversion method based on hybrid norm self-supervised learning, the problems of noise and outliers in multi-source mixed data are solved, achieving efficient signal separation and data quality improvement, and is suitable for complex noise environments.

CN120762095BActive Publication Date: 2026-05-26HARBIN INST OF TECH
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
HARBIN INST OF TECH
Filing Date
2025-07-03
Publication Date
2026-05-26

AI Technical Summary

Technical Problem

Existing technologies suffer from insufficient model stability when dealing with non-Gaussian noise, outliers, and data sparsity in multi-source mixed data, which limits the performance of deep learning methods in aliased signal separation.

Method used

An iterative inversion dealiasing method based on mixture norm self-supervised learning is adopted. By constructing a denoising convolutional neural network, using pseudo-separation to generate synthetic noise as a label, and combining iterative inversion with the projection gradient descent algorithm, the network is optimized to separate aliased signals.

Benefits of technology

It achieves accurate data separation in the face of complex aliasing noise, demonstrating good generalization ability and practical application value, and significantly improving acquisition efficiency and data quality.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120762095B_ABST
    Figure CN120762095B_ABST
Patent Text Reader

Abstract

This invention discloses an iterative inversion dealiasing method based on mixture norm self-supervised learning, belonging to the field of multi-source acquired aliased signal separation technology. It aims to address the issue of improving model stability when facing non-Gaussian noise, outliers, and data sparsity. The invention constructs a denoising convolutional neural network; it generates primary aliased data from the original seismic data through a pseudo-separation process, then generates secondary aliased data using an aliasing operator combined with the pseudo-separation process, and calculates the synthetic noise; it uses the primary aliased data combined with the synthetic noise to construct the input and labels of the denoising convolutional neural network; it constructs a loss function for the denoising convolutional neural network, resulting in a trained denoising convolutional neural network; it uses the trained denoising convolutional neural network as input to the original acquired data and outputs the denoised result; it uses the projection gradient descent algorithm to invert the denoised result, and the inverted result replaces the original seismic data. This process is repeated until the evaluation criteria are met. This invention exhibits good stability.
Need to check novelty before this filing date? Find Prior Art